All measurements described below were performed as part of the PROPHET-AMOS 2016 field campaign. Measurements were conducted throughout the month of July at the PROPHET facility at the University of Michigan Biological Station (UMBS) in northern Michigan (45.5588∘ N, 84.7145∘ W). The mixed deciduous and coniferous forest site consists primarily of isoprene-emitting species such as big-tooth aspen and red oak but also monoterpene-emitting species such as red maple, white pine, and paper birch (Ortega et al., 2007; Bryan et al., 2015). The site has been described in more detail elsewhere (Carroll et al., 2001; Ortega et al., 2007; Griffith et al., 2013). The majority of the measurements described below were performed near the top of the 31 m tower, approximately 10 m above the forest canopy by placing the instrument directly on the top of the tower, sampling from a glass manifold in the laboratory that pulled air from the top of the tower, or sampling from individual inlets from the top of the tower. Measurements of ozone were taken from the top of the nearby Ameriflux tower, which is 100 m to the north of the PROPHET tower. Table 1 summarizes the measurements used in this study.

During the campaign, isoprene, the sum of methyl vinyl ketone and methacrolein, monoterpenes, acetaldehyde, and other VOCs and oxygenated VOCs (OVOCs) were measured by the University of Minnesota using proton-transfer- reaction quadrupole-interface time-of-flight mass spectrometry (PTR-QiTOF) (Millet et al., 2018). In addition, C2–C10 alkanes and alkenes, butadiene, C6–C9 aromatic compounds, and isoprene were measured by IMT Nord Europe using a thermal desorption gas chromatography with flame ionization detection (GC-FID) instrument with a 1.5 h time resolution, while C2–C10 aldehydes, C2–C6 ketones, and C2–C4 alcohols were measured by thermal desorption GC-FID with mass spectrometry (GC-FID-MS) with a 1.5 h time resolution (Badol et al., 2004; Roukos et al., 2009). NO and NO2 were measured by the NCAR single-channel chemiluminescence instrument (Ridley and Grahek, 1990), ozone was measured by UV absorption by the University of Colorado and CO by laser-based off-axis integrated cavity output spectroscopy (Los Gatos Research) by the University of Houston and Rice University groups. Isoprene hydroxy hydroperoxides (ISOPOOH) were measured using a gas chromatograph chemical ionization mass spectrometry (GC-ToF-CIMS) instrument by Caltech (Vasquez et al., 2018). Isoprene hydroxy nitrates were measured by an iodine-adduct chemical ionization mass spectrometer by Purdue University (Xiong et al., 2015). Photolysis frequencies were measured using spectral radiometry (Shetter and Müller, 1999) by the University of Houston. Measurements of OH, HO2, and XO2 radicals are described in detail below.

Both OH and HO2 were measured using the Indiana University laser-induced fluorescence–fluorescence assay by gas expansion (IU-FAGE) instrument that has been described in more detail previously (Dusanter et al., 2009a; Griffith et al., 2013; Lew et al., 2020). Briefly, OH radicals are detected by laser excitation at 308 nm and subsequent resonant fluorescence detection. The sampled air expands into a low-pressure cell, which extends the OH fluorescence lifetime by reducing the concentration of species that may quench OH fluorescence and allows temporal filtering of OH fluorescence from more intense scattered laser light (Heard and Pilling, 2003).

The IU-FAGE laser system used in this study consisted of a Spectra Physics Navigator II YHP40-532Q that produced approximately 7.5 W of 532 nm radiation (10 kHz repetition rate) to pump a Sirah Credo dye laser (255 mg L-1 of Rhodamine 610 and 80 mg L-1 of Rhodamine 101 in ethanol), resulting in approximately 40 mW of radiation that is tunable near 308 nm. This laser system was housed in the laboratory at the bottom of the PROPHET measurement tower, and 308 nm radiation was focused onto the entrance of a 50 m optical fiber to transmit the laser emission to the sampling cell.

The IU-FAGE sampling cell was located atop the 31 m measurement tower, approximately 10 m above the forest canopy. Ambient air was drawn into the detection cell through a pinhole inlet (0.64 mm diameter) by means of three scroll pumps (Edwards XDS 35i) connected in parallel. The pumps were located at the bottom of the tower and connected to the sampling cell by two parallel 3.8 cm inner-diameter vacuum hoses, which resulted in a sampling cell pressure of 0.6 kPa (4.5 torr) and a flow of 3 SLPM through the sampling inlet.

On average, approximately 1.25 mW of 308 nm radiation exited the 50 m fiber and entered the sampling cell during the campaign. The laser emission enters the sampling cell perpendicular to the sampled air mass and intersects the expanded air in a White cell configuration with approximately 24 passes. The OH fluorescence is collected along an axis that is orthogonal to both the laser beam and sampled air mass and detected using a microchannel plate photomultiplier tube (MCP-PMT) detector (Hamamatsu R5946U), a preamplifier (Stanford Research Systems SR445), and a photon counter (Stanford Research Systems SR400). The MCP-PMT is turned off, and the photon counter is inactive during the laser pulse by means of a delay generator (Berkley Nucleonics 565) to allow the OH fluorescence to be temporally filtered from scattered laser light.

The net OH fluorescence signal is determined through successive spectral-modulation cycles in which the dye laser emission wavelength is tuned on and off resonance, with the Q1(3) transition of OH near 308 nm. A background signal, which primarily consists of scattered laser light that extends into the detection window, is established by tuning the laser emission off resonance with the OH transition and therefore not exciting OH radicals. This background signal is subtracted from the on-resonance signal. A reference cell in which OH is generated by the thermal dissociation of water vapor is used to ensure maximum overlap between dye-laser emission and the OH transition wavelength.

The IU-FAGE measurements of OH are subject to potential interferences when OH radicals are generated inside the detection cell. In the presence of water vapor, the photolysis of ozone by the laser can produce hydroxyl radicals through Reactions (R4) and (R5) (Davis et al., 1981a, b). R4O3+hv<340nm→O(1D)+O2R5O(1D)+H2O→OH+OH To characterize this and any other interference, a chemical scrubbing technique is used to remove ambient OH prior to entering the detection cell (Griffith et al., 2016; Rickly and Stevens, 2018; Lew et al., 2020). This chemical modulation technique is used to monitor levels of the laser-generated ozone-water interference and any other interference that may produce OH radicals inside the detection cell. Hexafluoropropylene (C3F6, 95.5 % in N2; Matheson Gas) was added through a circular injector 1 cm above the inlet with a flow rate of approximately 3.5 sccm to remove 95 % of external OH radicals (Rickly and Stevens, 2018). The differences between the measured OH during C3F6 addition and OH measurements including the interference represent the net ambient OH concentration in the atmosphere. The addition of C3F6 is modulated in between ambient OH measurements every 15 min for a duration of 10 min.

Measurements of HO2 were made indirectly after addition of NO to the sampled air mass to convert ambient HO2 to OH through the fast HO2 + NO → OH + NO2 reaction. A small flow (approximately 2 sccm) of NO (Matheson, 1 % in nitrogen) was added to the sampled air mass through a Teflon loop injector that was positioned directly below the sampling inlet, resulting in an added NO concentration of approximately 9×1011 cm-3. The fraction of HO2 converted into OH was measured during calibration experiments performed during and after the campaign and was 14.0±3.2 %. This low NO concentration minimized the impact of interferences from RO2 radicals derived from the OH-initiated oxidation of alkenes and aromatics that can be quickly converted to HO2 (Fuchs et al., 2011; Lew et al., 2018). The high conversion efficiencies reported for these RO2 radicals are due to the rapid decomposition of β-hydroxyalkoxy radicals that are formed from the RO2+ NO reaction. This decomposition forms a hydroxyalkyl radical that reacts rapidly with O2 to produce HO2 in the detection cell. This can lead to the detection of both HO2 and a fraction of RO2 radicals denoted as HO2∗ (HO2∗= HO2+αRO2, 0<α<1). Calibrations before and after the campaign similar to those described in Lew et al. (2018) indicated that the low NO concentration injected into the detection cell (approximately 9×1011 cm-3) resulted in an RO2-to-HO2 conversion efficiency of approximately 10 % for isoprene-based peroxy radicals and an RO2-to-OH conversion efficiency of less than 2 % (Fig. S1 in the Supplement). As a result, the HO2 measurements were performed at the low NO flow that effectively minimized the impact of any potential interference from isoprene-derived RO2 species that are dominant during the day at the PROPHET site (Griffith et al., 2013).

The instrument was calibrated by producing known concentrations of OH and HO2 from the photolysis of water vapor in air as described by Dusanter et al. (2008). The calibration source consists of an aluminum flow reactor with quartz windows on two opposite sides. Aluminum cartridges adjacent to each window house a low-pressure mercury pen lamp and a photodiode detector, both of which are continuously purged with dry nitrogen to stabilize the lamp temperature and prevent light absorption by atmospheric gases. Radiation from the mercury lamp passes through a bandpass filter centered at 185 nm prior to illuminating the flow reactor and detector. The location of the mercury lamp and photodiode is adjustable along the length of the calibration source to allow for the measurement of radical surface loss between the illuminated region and the exit of the calibrator. For calibrations during PROPHET, zero air was delivered to the calibration source at a flow rate of 50 L min-1. A variable fraction of the flow (5 %–40 %) was diverted through a set of custom bubblers containing high-purity water at the base of the tower. This humidified fraction of air was mixed back with the initial flow in approximately 35 m of PTFE tubing (1.25 cm i.d.) before entering the calibration source. Calibrations were performed before, after, and intermittently during the campaign to track changes in sensitivity (Dusanter et al., 2008). The uncertainty associated with this calibration technique is approximately 18 % (1σ) for both OH and HO2.

As previously mentioned, a 50 m fiber optic cable was used to transmit laser radiation to the sampling cell for the above-canopy measurements. The long fiber presented technical challenges that impacted the performance of the IU-FAGE instrument. Due to the length of the fiber, the laser pulse was temporally broadened prior to entering the detection cell and resulted in an increase of background laser scatter of the instrument. This broadened pulse made temporal filtering of scattered laser light difficult and ultimately led to a lower sensitivity and a higher limit of detection for OH. In addition, the length of the fiber corresponded to a decrease in transmission of radiation through the fiber. An average transmission of 8 % led to 0.76–2.15 mW of 308 nm radiation in the detection cell over the course of the campaign. Due to low laser power and high background signal, long averaging times were necessary for OH measurements. The limit of detection for OH was 6.5×105 molecules cm-3 (1σ, 2 h average). Measurements of HO2 were performed approximately once per hour with a limit of detection of 1.1×107 cm-3 (0.4 ppt) (1σ, 20 s average).

Total peroxy radicals were measured by the Ethane CHemical AMPlifier (ECHAMP) instrument that has been previously described in detail (Wood et al., 2017). This instrument is similar to traditional chemical amplifiers that mix ambient air with excess CO and NO (Cantrell and Stedman, 1982; Hastie et al., 1991; Cantrell et al., 1996) but instead utilizes chemical amplification by ethane (C2H6) and NO followed by detection of NO2 using cavity-attenuated phase-shift (CAPS) spectroscopy.

The ECHAMP inlet box was positioned on the top platform of the tower at a height of 31 m. Ambient air was sampled at a flow rate of 7.3 SLPM through a 0.4 cm inner diameter (ID) glass inlet that was internally coated with halocarbon wax to minimize radical loss on surfaces. A small flow (0.35 SLPM) of pure O2 was added through a side port to this main flow. The O2 addition increases the O2 mixing ratio in the sampled air to 24.6 % and reduces both the value and the variability of the relative humidity in the sampled air. The sampled air finally entered two reaction chambers at individual flow rates of 1.0 L min-1, with the remaining sampled air used to monitor temperature and RH. In the amplification chamber, the sampled air was immediately mixed with 20 sccm of 50 ppm NO and 20 sccm of 50 % C2H6 through an upstream reagent addition port, leading to final mixing ratios for NO and C2H6 of 1 ppm and 1 % respectively. A flow of 20 sccm N2 was added downstream, 100 ms later. In this chamber, ROx species are converted to HO2 and OH through reactions with NO (R6–R8). Reactions (R8)–(R12) repeat several times, leading to the formation of NO2 that is subsequently measured by a CAPS monitor. R6RO2+NO⟶RO+NO2R7RO+O2⟶HO2+productsR8HO2+NO⟶OH+NO2R9OH+C2H6⟶C2H5+H2OR10C2H5+O2+M⟶C2H5O2+MR11C2H5O2+NO⟶C2H5O+NO2R12C2H5O+O2⟶CH3CHO+HO2 In the background chamber, the sampled air was first mixed with NO and N2, and then C2H6 was added 100 ms later. In this mode, ambient radicals are removed by successive reactions with NO (R6–R8) until they form HONO via the OH + NO → HONO reaction, and therefore amplification chemistry does not occur. After reagent addition, air from each chamber enters identical CAPS monitors. The CAPS NO2 measurements from the background chamber represent ambient NO2, NO2 from the reaction of ambient O3 with added NO, and NO2 from reactions of ambient peroxy radicals with NO but not from ethane amplification reactions. The CAPS NO2 monitor following the amplification chamber measures the sum of that observed from the background chamber and NO2 produced from amplification chemistry. The amount of NO2 produced from amplification reactions (ΔNO2) is determined from the difference between the amplification and background chambers. The concentration of peroxy radicals is calculated by dividing [ΔNO2] by the experimentally determined amplification factor, F. R13ROx=ΔNO2(CAPSROx-CAPSOx)/F The amplification factor was determined as a function of relative humidity by producing known concentrations of peroxy radicals with two different calibration sources. The first source relies on the photolysis of water vapor method which is similar to that described above for the IU-FAGE instrument and is commonly used to calibrate other chemical amplifiers (Mihele and Hastie, 2000; Horstjann et al., 2014) and LIF-FAGE instruments (Heard and Pilling, 2003; Dusanter et al., 2008). This method produces equivalent concentrations of OH and HO2 that are quantified by O2 actinometry and measured concentrations of H2O and O3 in the calibration gas, and OH can be quantitatively converted to HO2 or isoprene peroxy radicals through the addition of H2 or isoprene, respectively. The second calibration source was based on the photolysis of methyl iodide (CH3I) at 254 nm to produce CH3O2 radicals (Anderson et al., 2019). The radical concentration is quantified by reaction with NO, in the absence of ethane, to produce NO2 that is measured by CAPS. During PROPHET the ECHAMP limit of detection was 1–3 ppt (2σ, 2 min average). Throughout the campaign, the CH3I calibration method was used as the primary source, and the water vapor photolysis method was used less frequently to quantify the relative response of ECHAMP to HO2 and CH3O2 radicals (see below).

As described in Wood et al. (2017) and Kundu et al. (2019), ECHAMP does not detect all peroxy radicals with equal sensitivity. A portion of RO2 radicals are converted to alkyl nitrites (RONO) and alkyl nitrates (RONO2) via association reactions with NO, and a portion of all sampled radicals are lost to wall reactions. Wall loss rate constants measured in the laboratory for halocarbon-coated 0.4 cm ID glass were typically 1.6 s-1 for HO2 at 60 % RH and less than 0.2 s-1 for CH3O2, with isoprene peroxy radical wall loss rate constants between those two values (Kundu et al., 2019). For the sampling conditions during PROPHET (7.3 SLPM flow rate, 13 cm inlet length), this suggests only 2 % of HO2 was lost to wall reactions. Furthermore, an expected 8 % of isoprene peroxy radicals are lost to formation of organic nitrates. The relative sensitivity of ECHAMP to HO2 radicals and CH3O2 radicals was quantified after the campaign by comparing its response to both types of radicals prepared at equal concentrations using the water vapor photolysis method. These measurements showed that the response to HO2 was 2 % lower than the instrument response to CH3O2. In the absence of sampling losses we would expect that the response to CH3O2 would be 10 % lower than the response to HO2 due to formation of CH3ONO (Wood et al., 2017). These results indicate that sampling losses of HO2 were more likely 10 % and almost equal to the loss of CH3O2 due to CH3ONO formation. Further details of a calibration source comparison between the LIF and ECHAMP instruments are provided in the Supplement.

Ambient concentrations of OH, HO2, and XO2 were modeled with the Master Chemical Mechanism (MCM) (Jenkin et al., 1997, 2015) and the Regional Atmospheric Chemistry Mechanism version 2 (RACM2) (Goliff et al., 2013). The RACM2 mechanism groups several species according to their reactivity and includes more than 350 reactions. While the near-explicit MCM is expected to better represent the complex oxidation chemistry of this environment, the grouped RACM2 model is more computationally efficient and simpler to use in a radical budget analysis. Due to the limited isoprene oxidation mechanism in the base RACM2 model, a series of reactions described by Tan et al. (2017) was incorporated based on the LIM1 mechanism proposed by Peeters et al. (2009, 2014). The resulting condensed version of LIM1 includes updated bulk reaction rate constants for the 1,6-H shift isomerization reactions of the isoprene peroxy radicals as parameterized by Peeters et al. (2014). These isomerization reactions lead to the formation of HO2 and hydroperoxyaldehydes (HPALDs), which can photolyze leading to OH production, and dihydroperoxy-carbonyl peroxy radicals (di-HPCARPs), which can rapidly decompose to produce additional OH radicals (Teng et al., 2017; Wennberg et al., 2018).

The Master Chemical Mechanism provides a near-explicit mechanism that describes the gas-phase chemical processes involved in the degradation of over 140 VOCs. Model simulations utilized both MCM version 3.2 and MCM version 3.3.1, the latter of which incorporates the explicit LIM1 mechanism and includes the equilibrium between different isoprene peroxy radical isomers and the H-shift isomerization reactions of specific isomers resulting in HOx radical recycling through the production of HPALDs as well as di-HPCARPs (Jenkin et al., 2015). In this mechanism, the equilibrium rate coefficients between different peroxy radical isomers were increased, and the 1,6 H-shift isomerization rate constants were decreased in order to match early experimental results of Crounse et al. (2014) (Peeters, 2015). These changes resulted in effective bulk 1,6 H-shift peroxy radical isomerization rate constants in MCM v3.3.1 that are approximately a factor of 5 lower than the original LIM1 recommended rates (Novelli et al., 2020).

Each of the chemical mechanisms were embedded in the Framework for 0-D Atmospheric Modeling (F0AM) (Wolfe et al., 2016) to calculate radical concentrations predicted by each mechanism. Modeled chemistry for both mechanisms was constrained to measurements of meteorological data and a wide variety of trace gas mixing ratios that were measured during the campaign (Tables S1 and S2). Model simulations were performed with a 30 min integration time and a 5 d spin-up to allow sufficient time to generate unmeasured secondary oxidation products. A 24 h lifetime was assumed for all calculated species to simulate loss via dry deposition and to prevent unexpected accumulation of some unmeasured species. Similar to Ren et al. (2013) and Lu et al. (2012), model sensitivity runs indicate that increasing this depositional loss by a factor of 2 results in changes of the modeled HO2 concentration of less than 4 %. Measurement constraints were synchronized to 30 min time intervals. Species that were measured more frequently were averaged to 30 min intervals, and linear interpolation was used for species measured with lower time resolution.

In cases when speciated measurements or complete measurement sets were not available for species important to radical chemistry, an appropriate correlation analysis or an average of previous measurements conducted at the PROPHET location was used to constrain the model. For example, measurements of the sum of methyl vinyl ketone and methacrolein (MVK + MACR) were available throughout the campaign from the University of Minnesota's PTR-QiTOF instrument, but speciated MACR was measured on some days by the IMT Nord Europe online GC-FID. An MVK:MACR ratio of 0.65:0.35 was derived from a correlation of the available measurements and used to constrain the model when speciated measurements were not available. Similarly, as the sum of monoterpenes was measured by PTR-QiTOF, the mixing ratio of total monoterpenes was constrained as α-pinene in model simulations. Because measurements of HONO concentrations at the top of the tower were unavailable, the model was constrained to the campaign average of previous measurements at this site (Griffith et al., 2013). Photolysis frequencies were calculated using a trigonometric parameterization based on solar zenith angle (Jenkin et al., 1997; Wolfe et al., 2016) and scaled according to measured values of J(NO2) to account for cloud coverage. The uncertainty of modeled radical concentrations is estimated to be 30 % based on uncertainties from model constraint inputs and the measured rate constants for each reaction (Griffith et al., 2013; Wolfe et al., 2016).

In addition to the standard RACM2 and MCM v3.2 models, and the expanded isoprene chemistry in RACM2-LIM1 and MCM v3.3.1, a third set of model simulations were conducted to investigate the influence of RO2+ RO2 accretion reactions and dimer formation on overall radical concentrations. A set of reactions were added to both RACM2-LIM1 and MCM v3.3.1 to create overall mechanisms (RACM-ACC and MCM-ACC) that incorporate RO2+ RO2 accretion reactions for both isoprene- and monoterpene-based peroxy radicals. Several studies have reported observations of highly oxidized C19-20H28-32O10-18 dimer products in chambers (Ehn et al., 2014) and in field measurements (Yan et al., 2016; Zha et al., 2018), suggesting that RO2 reactivity in the process of dimer formation increases along with functionalization and size of the RO2 radical (Berndt et al., 2018a, b). Rate constants for the added reactions were based on measurements from Berndt et al. (2018a, b) and are intended to represent complex autoxidation and dimer formation chemistry into a model process that results in net radical termination.

Rate constants for several RO2+ RO2 reactions used in this study are shown in Table 2. As described above, measurements of the sum of all monoterpenes were interpreted as α-pinene in the model, and thus rate constants measured in an exclusively α-pinene system (Berndt et al., 2018a) were used and provide only an estimation of the terpene chemistry at the PROPHET site that also includes emissions of β-pinene, limonene, and others (Carroll et al., 2001; Ortega et al., 2007). In addition to C10–RO2 radicals derived from monoterpenes, measured rate constants for self- and cross-reactions of C5-RO2 radicals derived from isoprene are also included, as well as more general, slower reactions between C10-RO2 and other smaller RO2 species. For the purposes of this study, rate constants are based on measurements of the least oxidized C10-RO2 species described in Berndt et al. (2018a) and thus may represent a lower limit in terms of autooxidation and dimer reactions as a radical sink. As such, the goal of this model was not to provide a detailed mechanism or exact representation of chemistry but instead to investigate the plausibility of autooxidation and dimer formation and the relative importance that the process may have as a radical termination process in a low-NOx, high-BVOC environment.

A complete suite of supporting measurements, including meteorological conditions and important chemical species that were used as model constraints is shown in Fig. 1, and campaign average measured values of important model constraints are shown in Fig. 2. In general, weather during the campaign was sunny with intermittent clouds, with some exceptions of more overcast days (8, 15, 17, and 24 July). Mixing ratios of NO, O3, and photolysis rate constants were similar to those observed during previous field campaigns at the same site (Griffith et al., 2013). The maximum observed NO mixing ratio was 480 ppt on 11 July, and the average peak mixing ratio of NO was approximately 115 ppt at 09:00 local time. NO mixing ratios at night were typically less than 0.5 ppt. Average ozone mixing ratios were between 25 and 35 ppb. Maximum average daytime temperatures of 24 ∘C were similar to measurements at this site in 2008 but warmer than measurements at this site in 2009, resulting in average mixing ratios of isoprene that peaked near 3 ppb at approximately 18:00, similar to that measured in 2008 but greater than that measured in 2009 (Griffith et al., 2013). These measurements are summarized along with those from previous campaigns at the PROPHET site in Table S3. Mixing ratios of anthropogenic VOCs were low, with average mixing ratios of toluene and benzene below 65 and 40 ppt, respectively.

Measured and modeled OH, HO2, and XO2 concentrations from 2 July through 31 July are shown in Fig. 3 with correlation plots shown in Fig. S2. Measurements of OH were hampered by high background signals and limited laser power. Diurnal profiles with a 2 h time resolution of the OH measurements are shown in Fig. 4, in addition to the model results. An average of all OH measurements performed during the campaign shows a peak of 1.25×106 molecules cm-3 at 13:00. Measurements of OH during the morning hours were significantly lower than all model calculations. An experimental OH budget based on measured concentrations of OH, HO2, and other species that contribute to OH production and loss is shown in Fig. S3. The imbalance between 07:00 and 12:00 suggests either a missing OH sink or errors with the OH measurement during this time. The reason for this discrepancy is not clear but may be the result of participant activity on the top of the tower near the detection cell in the mornings during the campaign which may have influenced the OH measurements, although a systematic measurement error during this time cannot be ruled out.

Measurements of potential interferences by chemical modulation through addition of C3F6 as described above did not reveal any significant unknown interferences, similar to that observed previously at this site (Griffith et al., 2013) but in contrast to measurements by the IU FAGE instrument during the IRRONIC (Indiana Radical, Reactivity and Ozone Production Intercomparison) campaign (Lew et al., 2020). Lew et al. (2020) found that the measured interference increased with both ozone and temperature, similar to that observed by Mao et al. (2012), who also measured a similar interference that increased with both temperature and total OH reactivity. Laboratory studies suggest that the interference could be due to the decomposition of Criegee intermediates inside the low-pressure detection cell leading to OH production (Novelli et al., 2014a, 2017; Fuchs et al., 2016; Rickly and Stevens, 2018), although estimated concentrations of Criegee intermediates in similar environments are too low to explain the observed interference (Novelli et al., 2017). Another proposed source of the interference is the decomposition of ROOOH molecules inside the FAGE detection cell formed from the reaction of OH with RO2 radicals (Fittschen et al., 2019). While the sources of these interferences are still unknown, one possible explanation for the absence of a measurable interference during PROPHET-AMOS is the lower measured mixing ratios of ozone and lower temperatures compared to those measured during IRRONIC, resulting in lower mixing ratios of isoprene and other BVOCs. Based on the observed correlation of the interference with ozone and temperature highlighted in Lew et al. (2020), it is likely that a similar interference was undetectable during PROPHET-AMOS.

The measured OH concentrations reported here are similar to previous measurements made by the IU-FAGE instrument at the PROPHET site in 2009 but are lower than those measured at the site in 2008, although the latter measurements suffered from poor precision (Griffith et al., 2013). The results reported here are also in contrast to measurements of OH at this site in 1998 as reported by Tan et al. (2001), who reported maximum daytime concentrations of approximately 4×106 cm-3 that were approximately a factor of 3 greater than model predictions (Table S3). While the mixing ratios of NOx and isoprene in 1998 were similar to those observed during PROPHET-AMOS, mixing ratios of ozone were higher in 1998, with the average maximum of approximately 45 ppb similar to that observed during the IRRONIC campaign (Lew et al., 2020). In addition, anomalously elevated concentrations of OH were observed at night in 1998 (Faloona et al., 2001; Tan et al., 2001). These results suggest that the OH measurements in 1998 at the PROPHET site may have been influenced by interferences similar to those observed by Mao et al. (2012), Feiner et al. (2016), and Lew et al. (2020). As illustrated in Fig. 4, the base RACM2 and MCM v3.2 models are able to reproduce the maximum observed OH concentrations to within the combined measurement precision and uncertainty of the models. The addition of LIM1 chemistry to the models increased the predicted average maximum OH concentration by approximately 20 % between MCM v3.2 and MCM v3.3.1 and by 30 % between RACM2 and RACM2-LIM1, with the MCM v3.3.1 maximum modeled OH concentrations approximately 30 % greater than the measured concentrations and the RACM2-LIM1 maximum modeled OH concentrations approximately 60 % greater than the measured concentrations. These results are in contrast to several previous LIF measurements in forested environments (Rohrer et al., 2014), in which measured OH concentrations were significantly higher than modeled predictions. However, the results reported here are similar to those found by Feiner et al. (2016) in an Alabama forest during SOAS (Southern Oxidant and Aerosol Study) where isoprene was the dominant BVOC. In that study, the modeled OH concentrations using MCM v3.3.1 were in good agreement with the measured concentrations when interferences were subtracted from the measurements.

While the predictions by both the RACM2 and MCM models are within the combined uncertainty of the measurements and the models, the MCM v3.3.1 results are in better agreement with the measurements (Fig. 4), which could suggest that the measurements are consistent with the lower effective bulk 1,6-H shift peroxy radical isomerization rate constants in MCM v3.3.1 compared to the original LIM1 recommended rates (Novelli et al., 2020). This is in contrast to the results of from the IRRONIC campaign discussed above, where the MCM v3.3.1 model underpredicted the measured concentrations by approximately a factor of 2, with the RACM2-LIM1 model predictions in better agreement with the measurements (Lew et al., 2020). Similarly, Novelli et al. (2020) reported that the MCM v3.3.1 mechanism underpredicted measurements of OH by a factor of approximately 1.4 during isoprene oxidation experiments in the SAPHIR (Simulation of Atmospheric Photochemistry In a large Reaction) chamber when mixing ratios of NO were less than 0.2 ppb. Unfortunately, the poor precision of the OH measurements reported here do not allow a robust test of the two mechanisms.

The time series of measured and modeled HO2 and total peroxy radical (XO2) concentrations from 2 July through 31 July are shown in Fig. 3, and correlation plots of the measured concentrations and MCM v3.3.1 model predictions are shown in Fig. S2. Daily maxima were typically observed between 13:30 and 15:30 local time and ranged from 1.7×108 cm-3 (6.7 ppt) to 7.0×108 cm-3 (28.2 ppt) for HO2 and 2.7×108 cm-3 (10.8 ppt) to 1.3×109 cm-3 (52.1 ppt) for XO2. Measured diurnal average profiles are shown in Fig. 4 along with average model results that consider only the days on which each respective species was measured (7/3–7/9 and 7/11–7/31 for HO2; 7/9–7/14 and 7/18–7/26 for XO2). In addition, measured RO2 mixing ratios (HO2 measured by LIF subtracted from XO2 measured by ECHAMP) are compared with modeled RO2 mixing ratios in Fig. S4. The average maximum (12:00–15:00) of HO2 measurements performed during the campaign was 2.85×108 cm-3 (11.6 ppt), while the maximum daytime average of the XO2 measurements was approximately 7.7×108 cm-3 (29.0 ppt).

The measured HO2 concentrations were similar to previous measurements at this site. Median daytime maximum concentrations of HO2* measured in 2008 were approximately 7×108 cm-3 (28 ppt), while median daytime maximum concentrations of HO2* measured in 2009 were approximately 5×108 cm-3 (20 ppt), with nighttime concentrations below 1×108 cm-3 (4 ppt) during both years (Griffith et al., 2013). The conversion efficiency of isoprene peroxy radicals to the measured HO2* concentrations during these studies was estimated to be approximately 90 %, suggesting that the measured HO2* concentrations reflected the sum of HO2+ isoprene peroxy radicals. Given that isoprene peroxy radicals contribute to approximately 33 % of the total peroxy radical concentrations during the daytime, the measured HO2* concentrations in 2008 and 2009 were greater than HO2 but less than XO2 concentrations (Table S3). When compared to 2009, the higher HO2* concentrations measured in 2008 were likely due to the higher mixing ratios of HCHO observed in 2008, leading to greater rates of radical production (Griffith et al., 2013). The higher mixing ratios of HCHO may be a result of the higher mixing ratios of isoprene leading to a greater production of HCHO during the warmer temperatures observed in 2008 (Griffith et al., 2013).

Average daytime maximum concentrations of HO2 measured at this site in 1998 were reported to be approximately 16 ppt 3.9×108 cm-3 (16 ppt), with nighttime concentrations less than 1.2×108 cm-3 (5 ppt) (Tan et al., 2001), similar to the measurements in this study. However, it is not clear whether the 1998 HO2 measurements were influenced by interferences from isoprene-based peroxy radicals, as discussed above. As a result, these measurements may be an upper limit to the actual HO2 concentrations. The measured XO2 concentrations are similar to the total RO2+ HO2 concentrations measured at this site in 1997 by Mihele and Hastie (2003), who reported daytime maximum mixing ratios between 20 and 65 ppt using a radical chemical amplifier technique, and nighttime mixing ratios of 3–6 ppt. Although not measured at the site during the 1997 campaign, monoterpene mixing ratios observed in 1998, 2008, and 2009 were similar to measurements from 2016 (Table S3).

As illustrated in Fig. 4, the base RACM2 and MCM v3.2 models overpredict both the measured HO2 and XO2 concentrations during the daytime, although the agreement with the measured HO2 concentrations is within the combined uncertainty of the measurements and the model. The base RACM2 model overpredicts the measured average maximum HO2 concentrations by approximately 30 %, while the MCM v3.2 overpredicts the measured daytime maximum HO2 by approximately 60 %. However, including the LIM1 isoprene oxidation mechanism increases the daytime HO2 concentrations predicted by the base models by approximately 15 % and 35 % for MCM and RACM2 models, respectively (Fig. 4). Overall, both the RACM2-LIM1 and MCM v3.3.1 models overpredict the measured daytime maximum HO2 concentrations by approximately 80 %, which is outside of the combined measurement and model uncertainties. Similarly, the base RACM2 and MCM v3.2 models as well as the updated RACM2-LIM1 and the MCM v3.3.1 models overpredict the daytime XO2 concentrations by more than a factor of 2, with predicted daytime maximum XO2 mixing ratios ranging from 65.5 (RACM2) to 72.6 (RACM2-LIM1) ppt (1.6–1.8×109 cm-3).

The model overprediction of the daytime measured HO2 concentrations is consistent with model simulations of the measured HO2* concentrations at this site in 2008 and 2009 (Griffith et al., 2013). In 2008, a base RACM model overpredicted the measured HO2* concentrations by approximately 30 % on average, while the same model overpredicted the HO2* concentrations measured in 2009 by approximately 50 %. Similar to the results presented here, addition of the LIM1 mechanism for isoprene oxidation to the RACM model likely would have increased the discrepancy between the 2008 and 2009 measurements. However, these model results are in contrast to those observed at this site in 1998, where a RACM-based model was able to reproduce the reported measured HO2 concentrations (Tan et al., 2001). As discussed above, these measurements likely represent an upper limit to the actual HO2 concentrations as it is not clear whether the measurements of HO2 were free from interferences from isoprene-based and other alkene-based peroxy radicals (Fuchs et al., 2011; Lew et al., 2018). As a result, it is likely that this RACM-based model of Tan et al. (2001) overestimated the actual HO2 concentrations in 1998. The results reported here are also in contrast to the results of Mihele and Hastie (2003), who found that a 0-D MCM-based model could reproduce the measured daytime XO2 concentrations on several days. Similarly, these results are in contrast to the IRRONIC campaign, where the MCM and RACM2 models were able to reproduce the measured HO2* concentrations to within 30 % (Lew et al., 2020), and the results from SOAS, where the MCM v3.3.1 model was able to reproduce the measured HO2 concentrations to within the combined uncertainties of the measurement and the model (Feiner et al., 2016). The ability of the models to reproduce the measured peroxy radical concentrations in these studies may reflect the higher mixing ratios of NOx observed at the PROPHET site in 1997, the SOAS site, and at the IRRONIC site, resulting in a greater contribution of the RO2+ NOx reactions to the fate of peroxy radicals during these campaigns (Mihele and Hastie, 2003; Sanchez et al., 2018; Lew et al., 2020). An analysis of the discrepancies between the PROPHET and IRRONIC campaigns will be presented in a subsequent paper.

The composition of the total peroxy radical concentration (XO2) in the RACM2-LIM1 model is shown in Fig. 5. As illustrated in this figure, the model predicts that HO2 radicals comprise approximately 33 % of the total daytime maximum XO2 concentration, with isoprene peroxy radicals accounting for approximately 33 %, methyl peroxy radicals for approximately 12 %, and acyl peroxy radicals for approximately 7 % and with peroxy radicals from alkane, alkene, and terpene oxidation comprising the remaining 15 %. Given that the model agreement with the measurements is better for HO2, the majority of the discrepancy between the modeled and measured XO2 concentrations is likely due to a greater overestimation of RO2 radicals, including isoprene-based peroxy radicals. These results are in contrast to that reported by Kundu et al. (2019), who found that their measurements of XO2 concentrations by the ECHAMP instrument during the IRRONIC campaign could be reproduced on several days by a model incorporating the MCM v3.2. As discussed above, the ability of the models to reproduce the measured concentrations during IRRONIC may reflect the higher mixing ratios of NO observed during this campaign, resulting in a greater contribution of RO2+ NOx reactions to the fate of peroxy radicals at this site.

During the nighttime, the models reproduce the measured HO2 concentrations but overestimate the measured XO2 radical concentrations (Fig. 4). The RACM2 and MCM models overpredict the nighttime XO2 concentrations by a factor of approximately 4, with the RACM2-LIM1 model predicting mixing ratios of total peroxy radicals of approximately 27 ppt between 21:00 and 06:00 and the MCM v3.3.1 model predicting XO2 mixing ratios of approximately 36 ppt during the night (Fig. 4) compared to the measured concentrations of less than 10 ppt. These results are similar to those from the 1997 PROPHET campaign in which measured XO2 mixing ratios of 3–6 ppt were overpredicted by more than a factor of 10 by a model that included reactive terpene emissions (Mihele and Hastie, 2003; Sillman et al., 2002). This is in contrast to the results of Kundu et al. (2019), who found that the MCM v3.2 could reproduce measured nighttime mixing ratios of less than 10 ppt during the IRRONIC campaign, which is likely a result of the elevated NOx mixing ratios at that site.

The RACM2-LIM1 model predicts that approximately 50 % of the nighttime total XO2 radical concentration is composed of peroxy radicals derived from the ozonolysis of monoterpenes (Fig. 5). As mentioned above, the measured sum of monoterpenes was constrained as α-pinene in all model simulations, resulting in an average monoterpene ozonolysis rate constant that is likely similar to that expected from previous speciated measurements of monoterpenes, including limonene and β-pinene, at this site (Ortega et al., 2007; Kim et al., 2011). However, the average ozonolysis rate constant assumed in the model could represent an upper limit if the monoterpene composition was dominated by species less reactive with ozone (e.g., camphene, cymene) or a lower limit if more reactive terpene species were present (e.g., ocimene, limonene) (Atkinson et al., 1990; Khamaganov and Hites, 2001; Atkinson and Arey, 2003).

The addition of the RO2+ RO2 accretion reactions described above to the RACM2-LIM1 and MCM v3.3.1 models (RACM-ACC and MCM-ACC) significantly reduces the predicted XO2 radical concentrations at night by 50 %, lowering the measurement–model discrepancy to less than 5 ppt for the RACM-ACC model. As shown in Fig. 5, this is largely due to a reduction in the concentration of organic peroxy radicals derived from monoterpenes due to the relatively large rate constants for the associated RO2+ RO2 accretion reactions (Table 2).

A radical budget analysis for OH, HO2, isoprene-based peroxy radicals (ISOP) and total ROx was conducted to provide information about the processes that drive radical production and the radical loss pathways in this environment and also to highlight the relative importance of the changes in radical chemistry upon the addition of the LIM1 mechanism and accretion reactions. Figure 6a illustrates the campaign average production and loss pathways of OH for the RACM2-LIM1 model. This includes both initiation reactions and propagation steps that produce OH in blue, while termination pathways are shown alongside propagation steps that convert OH to HO2 or RO2 in red. The addition of the LIM1 reactions increases the maximum OH production rate at 13:45 by 35 % from 2.01 ppb h-1 in RACM2 to 2.71 ppb h-1 in RACM2-LIM1, primarily due to the isomerization of isoprene peroxy radicals to form HPALDs, which readily photolyze to form OH, and also di-HPCARPs, which rapidly decompose to produce additional OH radicals (Peeters et al., 2014; Teng et al., 2017; Wennberg et al., 2018). In the morning (06:45–13:15), RACM2-LIM1 suggests the HO2+ NO reaction is the dominant source of OH radicals, accounting for 41 % of total OH production and as much as 53 % when NO mixing ratios are the highest. This decreases to 21 % in the afternoon and evening as the NO concentration decreases. Photolytic processes are significant throughout the day, with ozone and HONO photolysis contributing up to 28% and 13 % respectively during the day. Ozonolysis of alkenes, primarily monoterpenes, is a minor contributor of up to 6 % during the day but is the dominant source of OH at night.

Reaction with isoprene is the dominant loss pathway for OH radicals accounting for approximately 60 % of daytime OH reactivity. Other VOCs (16 %), namely monoterpenes, and OVOCs (10 %), such as formaldehyde, methyl vinyl ketone, and methacrolein, make up the majority of the remaining daytime OH reactivity. Propagation through reaction with CO is minor (6 %), and termination through the OH +  NO2 reaction is not significant (< 2 %). Consistent with the OH radical budget, the OH + isoprene reaction is the dominant source of isoprene-based peroxy radicals (ISOP; Fig. 6b), with the ISOP + NO reaction accounting for approximately 53 % of their total loss in the morning, while the ISOP + HO2 reaction and peroxy radical isomerization reactions in the LIM1 mechanism account for 62 % of isoprene-based peroxy radical loss in the afternoon. The ISOP accretion reaction accounts for only 8 % of the loss of isoprene-based peroxy radicals in the afternoon.

Figure 6c illustrates the campaign average HO2 radical production and loss pathways for RACM2-LIM1. The production of HO2 in the RACM2-LIM1 model is largely due to turnover from the RO2+ NO reaction. During the morning, when NO concentrations are greatest, 32 % of HO2 is produced from the reaction of NO and peroxy radicals derived from isoprene, while 17 % is produced from the reaction of NO with other RO2 species. In addition to reactions with NO, the photolysis of formaldehyde can account for up to 15 % of daytime HO2 production and turnover from the OH + CO reaction near 8 %. HO2 loss is primarily due to reaction with NO in the morning (48 %) but dominated by the HO2 self-reaction, reaction with isoprene RO2 to form ISOPOOH, and reaction with other peroxy radicals in the afternoon and evening (21 %, 30 %, and 16 % respectively).

The total ROx radical budget is illustrated in Fig. 6d. The addition of LIM1 reactions increases the maximum radical initiation rate by 28 % from 2.11 to 2.69 ppb h-1, again primarily due to fast photolysis of HPALDs and decomposition of di-HPCARPs. Overall radical initiation in RACM2-LIM1 is largely due to photolytic processes, with a combined 51 % from ozone photolysis (26 %), HONO (13 %), and HPALDs (18 %) and 32 % from the photolysis of other species such as hydrogen peroxide, aldehydes, organic peroxides, and nitric acid. Ozonolysis is a consistent radical initiation source of approximately 0.21 ppb h-1 throughout the day, which dominates ROx initiation at night and is a significant contributor (11 %) throughout the day when photolysis sources are dominant.

Daytime termination of radicals in RACM2-LIM1 is dominated by peroxy radical self- and cross-reactions, primarily the reaction of isoprene peroxy radicals with HO2 to form ISOPOOH (44 %) but also the HO2 self-reaction (16 %) and HO2+ other RO2 species (21 %). Radical reactions with NOx were less significant due to the low NOx concentrations and accounted for at most 0.13 ppb h-1 or less than 8 % of the RACM2-LIM1 termination budget when NOx mixing ratios were highest and less than 3 % total during the daytime. The addition of RO2+ RO2 accretion reactions in the RACM-ACC model provides an alternative pathway that results in a termination rate equivalent to half that of HO2+ RO2 reactions at night (4.0×105 molecules cm-3 s-1) and accounts for 30 % of total ROx termination during this time. During the day, when isoprene and NO mixing ratios are higher, these reactions only contribute to 9 % of the overall termination due to the lower rate constants for reactions of C5-RO2 from isoprene (Table 2). As shown in Fig. 4, this results in better agreement between the measurement and model at night, but model overprediction during the day remains.